ChatGPT Traffic Drop in GA4: Bug or Trend? | Refine Labs

Mid-October my LinkedIn feed was basically one topic. ChatGPT traffic falling off a cliff, GA4 screenshots, people calling it the end of AI search. Probably an overreaction.

We saw a drop too. My first instinct was to pull the data and post something that week. I’m glad I didn’t. We waited until we had all of October. Partly to see if it was a bug or something real. Partly because we only get somewhere between 1 and 17 ChatGPT sessions a day, and at that volume a week of data doesn’t tell you much of anything. Where I ended up: it was mostly the bug, and most of the drop has come back. Not all of it though. And I can’t tell you from GA4 why.

What happened

Quick recap if you missed it. After GPT-6 rolled out (paid Oct 7, free Oct 8), links inside ChatGPT answers stopped carrying the utm_source=chatgpt.com tag and the referrer. GA4 had nothing to go on, so those visits showed up as Direct.

A few of the numbers that were going around:

• Adobe’s Cedric Huesler looked at about 500 pages where ChatGPT drove 5%+ of visits. Oct 9–13 vs. the week before: ChatGPT down about 31,000 visits a day, Direct up about 26,000 a day on the same pages. (LinkedIn, via Search Engine Roundtable)
• Graphite saw ChatGPT-attributed sessions in GA4 drop 22% to 57% across the sites it tracks, starting Oct 9. (Search Engine Journal)
• LLM Pulse saw ChatGPT sessions down about 46% on Oct 9–11 while total traffic stayed flat. (LLM Pulse)
• Profound found the share of ChatGPT answers with at least one tagged link fell from about 98% on Oct 8 to 66% by Oct 14. (Profound)

Turned out to be a bug. Huesler says OpenAI confirmed it on Oct 14, and by the next morning tagged clicks were back to roughly 95–100% of normal on the sites he tracks. (Search Engine Journal)

So, fixed. What I wanted to know was what our numbers looked like after.

October was a dip, and most of it came back

For context, July through September we averaged around 9.7 identified ChatGPT sessions a day. The first two weeks of October we were at about 5.

Then it started climbing back. 6.6 a day the week of Oct 19. 8.1 the week of Oct 26. If I’d posted on Oct 15, I would’ve told you we were down 45%. Two weeks later most of that was gone. That’s kind of the whole lesson for me. One thing I can’t explain - the dip started a few days before the bug. Could just be a slow week. When you’re at 3–9 sessions a day, a slow week and a tracking bug look exactly the same.

But we’re not all the way back

October overall averaged 6.1 a day. Last October was 10.3. Even our best recent week is under the summer average.

Is something softening? Maybe. I wouldn’t call it a trend off one month at this volume. November will help.

What bugs me more is that even if it is real, GA4 can’t tell me why. ChatGPT could be mentioning us less. People could be getting what they need in the chat and not clicking through. Or some visits are still losing attribution somewhere -- the app, ChatGPT’s browser, untagged links. Each of those would mean doing something different about it, and the referral report looks identical for all of them.

Direct didn’t absorb it

The usual check for hidden ChatGPT traffic is whether Direct goes up when ChatGPT goes down.

During the bug week it didn’t. Both dropped. Then late October, Direct jumped while ChatGPT was still low, which honestly looked a lot like what everyone was posting about.

Look at the size of the bars though. Compared to the summer, ChatGPT was down 2–4 sessions a day in October. Direct was up as much as 105. You could hand every missing ChatGPT visit to Direct and you’d barely see it.

That late-October jump is our acquisition news. We saw the same thing, smaller, after my promotion announcement in September. People click through from LinkedIn, the LinkedIn app drops the referrer a lot of the time, and it lands in Direct.

I’ve always thought of Direct as a junk drawer. It’s typed URLs and bookmarks, sure. It’s also Slack links, email apps, LinkedIn’s app, and for a week in October, ChatGPT. If something else happened that month, you can make Direct say pretty much whatever you want.

ChatGPT sends us people checking us out

This was the part I didn’t expect.

Over the last 90 days, 58% of our identified ChatGPT visits landed on the homepage. Another 18% went to pages about the company -- our framework page, Chris Walker’s page, customer stories, pricing, careers. The blog got about 13%.

I went in assuming this was a content story. It doesn’t look like one. It looks like people asking ChatGPT about Refine Labs, or about agencies like us, and then clicking over to check us out. I can’t prove that from GA4, so it’s a guess. But it lines up with what Tomek Rudzki at Peec AI saw after GPT-6, where 54% of the direct links ChatGPT put in answers pointed to homepages. (Search Engine Journal)

This is where I think a lot of B2B teams have it backwards. AI visibility gets treated like an SEO project. From what we’re seeing it’s a lot closer to brand. What does ChatGPT say about you when a buyer asks, and does your homepage hold up when they get there?

If you’re trying to read your own numbers

If your ChatGPT traffic is in the single or low double digits a day, here’s what I’d actually do.

• Give it a month before you react, and compare against the last few months and the same month last year. Not just the week before.
• Find the pages ChatGPT sends people to and watch Direct on those, not sitewide. Skip the homepage, it gets too much Direct to tell you anything.
• Write down what else was going on. Launches, announcements, a big LinkedIn post. It all shows up in Direct.
• Check what ChatGPT is actually saying about you. Run the prompts a buyer would ask every couple weeks, or use a tool for it. That’s the only way I know to separate “cited less” from “tracked less.”
• Look at pipeline before anything else. If meetings and opps didn’t move, it’s probably a measurement problem.

And annotate Oct 8–15 in GA4 so nobody panics about it six months from now.

Where I landed

I figured this post would be either “the bug hit us” or “it skipped us.” It’s neither, really. Most of the drop was temporary. Direct didn’t soak it up. We’re still a little under where we were, and I don’t know why yet. I’ll pull November and see if it holds. Curious if anyone else is seeing the same thing - especially B2B teams at our kind of volume.

How we pulled this

Refine Labs GA4, January 2025 – November 1, 2026. ChatGPT = sessions where session source contains “chatgpt” (all came through as chatgpt.com). Direct = session source (direct), sitewide. Weekly figures are average sessions per day. We excluded March–May 2026 because of a separate tracking issue on our site. The landing page breakdown covers Jul 17 – Oct 14, 2026.

This is one website, not a controlled study. It can’t see ChatGPT visits that arrived with no tag or referrer.

Sources

• Search Engine Journal: ChatGPT Traffic Drop Appears To Be A Tracking Problem
• Search Engine Land: ChatGPT appears to drop UTM tracking
• Search Engine Roundtable: OpenAI Drops UTM Parameters From ChatGPT Results
• Profound: ChatGPT’s missing UTM tags
• LLM Pulse: ChatGPT Referral Traffic Fell by Almost Half on October 9

‍